Latest News

Smart Toilets, AI, and Other Efforts to Beat Colorectal Cancer

By Paul Nicolaus

June 9, 2020 | Colorectal cancer (CRC) is one of the most common forms of cancer and one of the most common causes of cancer death. In recent months, newly published findings on topics such as family history and the role of the gut microbiome continue to shed light on this disease, and emerging tools like AI-based screening and a “smart toilet” point toward new possibilities.

There were 1.8 million cases of CRC across the globe in 2018, according to the World Health Organization, trailing only lung and breast cancer in terms of sheer numbers. This year, there will be nearly 150,000 newly diagnosed cases and over 50,000 deaths caused by this disease in the US alone, according to American Cancer Society estimates.

CRC incidence rates in the US have generally been declining in recent decades thanks to changing patterns in risk factors like smoking and improved screening. In younger adults, however, rates have been on the rise, increasing since the mid-1980s in adults between the ages of 20 and 39 and since the mid-1990s in those between 40 and 54.

A similar pattern has emerged in other parts of the world. CRC prevalence in adults under the age of 50 increased in 19 countries, according to one study, and in eight other countries (Australia, Canada, Denmark, Germany, New Zealand, Slovenia, Sweden, and the UK) that had stable or declining trends in older adults.

While the reasons behind this trend aren’t yet clear, this could be associated with changes in risk factors ranging from a more sedentary lifestyle to less healthy eating habits.

Starting screening at an earlier age based on family history is one of the main recommended strategies for the prevention and detection of early‐onset CRC. Several medical societies suggest screening starting at 40 years of age (or 10 years before the age at diagnosis of the youngest relative diagnosed with CRC) for individuals with a first-degree relative with CRC.

Even so, the data supporting the effectiveness of this approach are limited, so Samir Gupta of the VA San Diego Healthcare System and the University of California San Diego and colleagues took a closer at the possible impact of family history-based guidelines for screening by examining information on individuals 40 to 49 years of age—2,473 with CRC and 772 without.

Published Apr. 20 in CANCER (doi: 10.1002/cncr.32851), the study revealed that among the people who presented with colon cancer between the age of 40 and 49, about 1 in 4 met the criteria for family history-based early screening. When those family histories were analyzed in greater detail, the researchers found that nearly all those who met family history-based criteria for early screening could have had a recommendation for initiating their screening sooner than the age at which they actually presented with cancer.

“I think there’s a glass half full, half empty story here,” Gupta told Diagnostics World. The study shows that a quarter of those who presented with young-onset cancers have a family history. “If we act on that, we may really have a chance of diagnosing that cancer or possibly even preventing it if we can catch it at the stage when it’s still a polyp and hasn’t developed yet into cancer,” he said.

On the other hand, 75% of those who presented with young-onset cancer in this study would not have been flagged to start screening early based on family history. “What that means is that we need to continue developing strategies to identify which people need to start screening early,” Gupta added.

AI and Smart Toilet Technology

Other newly published papers have detailed efforts to develop tools that could help improve the detection of this disease.

A team led by Saeed Hassanpour, associate professor of Biomedical Data Science and Epidemiology at Dartmouth’s Geisel School of Medicine, has created an AI model that classifies colorectal polyps on histology slides, for instance. In a prognostic study, he and colleagues examined the classification of the four most common polyp types using a set of internal slides as well as slides from 24 different US institutions.

The findings, published Apr. 23 in JAMA Network Open (doi:10.1001/jamanetworkopen.2020.3398), reveal that the model performed with the same level of sensitivity and accuracy as practicing pathologists. For the internal evaluation on 157 slides, it achieved a mean accuracy of 93.5% compared with local pathologists’ accuracy of 91.4%. On the external test set of 238 slides, the model had an accuracy of 87% compared with local pathologists’ accuracy of 86.6%.

The significance of his team’s experiment, Hassanpour told Diagnostics World, is it revealed that a model built and trained on a data set from one medical center could “achieve high accuracy and good efficacy on data coming from a diverse set of hospitals and institutions.”

The majority of CRC cases arise from polyps, he explained, which are growths in the lining of the colon or rectum that can progress to cancer. If a patient undergoes colonoscopy and the polyps are resected, cancer can be prevented, but the tricky thing about polyps is that they reoccur. After the screening, patients need to undergo surveillance for recurrence. The characterization and classification of these polyps is a major factor in patient management, and so is deciding upon the frequency of colonoscopies and follow-ups.

Several years ago, while discussing the potential of using AI in this domain, Hassanpour learned that there is plenty of variability among pathologists regarding the classification of colorectal polyps—even among GI pathologists, who are experts in this domain. His team’s efforts to develop an AI-based tool focuses on the notion that it could help reduce this variability and improve accuracy, all while enhancing the efficiency of pathologists.

Moving forward, the group is working on a clinical trial to explore the use of the algorithm as a CRC screening tool. The trial is run in a pathologist plus AI clinical team environment to compare the performance of pathologists using this tool to those not using it while reading polyp slides. “We integrated this AI model with a visualization and easy-to-use graphical user interface and we were in the midst of a clinical trial—we actually completed the first arm of this study,” Hassanpour said, but that work was put on pause due to COVID-19.

The end goal is to create a software application that can help pathologists carry out their work even more effectively. Although confirmation of diagnoses would still be required by human experts, the AI model could help triage slides and may help improve basic access to these types of services—especially in rural settings.

Meanwhile, Sanjiv “Sam” Gambhir and an international group of co-authors from the United States, Canada, South Korea, and the Netherlands presented proof-of-concept “smart toilet” technology. It can detect an array of disease markers in urine and stool, including those of some cancers, such as CRC. The technology, detailed in a paper published Apr. 6 in Nature Biomedical Engineering (doi: 10.1038/s41551-020-0534-9), makes use of an ordinary toilet that includes a variety of technologies inside the bowl.

Urine and stool samples are captured on video and processed by a set of algorithms that can distinguish normal urodynamics (like flow rate, stream time, or total volume) and stool consistencies from those that are unhealthy. In addition to physical stream analysis, urinalysis test strips mounted within the toilet system automatically interact with the urine stream using a motion sensor to measure molecular features.

Like smart watches, the technology falls into the category of continuous health monitoring, but it has its advantages. Unlike wearables, you can’t remove a smart toilet, Gambhir, a Stanford professor and chair of radiology, said in a news release. “Everyone uses the bathroom—there’s really no avoiding it—and that enhances its value as a disease-detecting device,” he added.

It’s still early, considering only about 20 participants have tested the toilet over several months, but the vision is big. The developers see it as a technology that could become part of a typical home bathroom. It was designed as an add-on, similar to a bidet, that could be mounted onto an existing toilet and is not intended to replace a doctor or a diagnosis. Ideally, an app would notify the user’s health care team, which would enable medical professionals to decide on any next steps to make a proper diagnosis.

As Gambhir and colleagues continue to develop the technology, they are looking to grow the number of participants and integrate molecular features into stool analysis.

Exploring the Gut Microbiome’s Role

Researchers at the University of Michigan are finding new clues about how the composition of the gut microbiome could contribute to the development of CRC. The group discovered that genetically identical mice from two different research colonies had very different susceptibilities to CRC when exposed to a carcinogen and an agent that promotes gastrointestinal inflammation, according to findings published Apr. 7 in Cell Reports (doi: 10.1016/j.celrep.2020.03.035).

The researchers hypothesized that differences in the gut microbiota between these two groups of mice were contributing to the difference in tumor risks. This prompted a series of microbiome transfer experiments that sought to determine which microbes are more important than others in terms of dictating disease risk. The researchers also took the pups of one group of mice and fostered them with mothers of the other group to see whether it is possible to transfer the microbiota and the phenotype of either increased or reduced tumor risk in that manner.

“And surprisingly, we’ve never found a complete transmission of phenotype or of the microbiota,” Grace Chen, associate professor of hematology/oncology at Michigan Medicine and member of the Rogel Cancer Center told Diagnostics World. “We always ended up with some sort of hybrid microbiome when we did these types of transfer experiments.”

They next tried to cross-reference the microbiome composition of each of the mice with how many tumors they developed and used statistical methods to pinpoint specific microbes of interest, she explained, which is how they arrived at their list of microbes (detailed in the paper) that consistently correlated with either high or low numbers of tumors. “A lot of these, unfortunately, are uncharacterized or have not been previously sequenced or grown in culture by any other research group,” Chen said. Currently, her team is in the process of trying to isolate these microbes to see whether they will either reduce or increase tumor susceptibility.

An important takeaway from these studies, according to Chen, is that they couldn’t find one individual bacterium that was responsible for the tumor phenotype. “It was more of a consortium of bacteria.”

One limitation of the research, she acknowledged, is the focus on the microbiota of the mouse. Although it has plenty of similarities with the human microbiota, there are distinct differences. A critical aspect that the researchers need to follow up on is whether any of the microbes that they identified in the mice that seem to be predictive of colon tumor risk are also found in humans.

Moving forward, Chen and colleagues are looking at other mouse models that developed spontaneous colon tumors—not necessarily from inflammation, but from underlying genetic mutations.

Another aspect they want to explore further centers on T cells, which play an important role in fighting cancers and killing tumor cells. Their study results revealed there were more T cells in the colon tissue of mice with bacteria from the second colony (which developed more tumors and had a more significant inflammatory response).

It’s a bit counter-intuitive considering T cells are typically associated with improved outcomes in CRC patients, but Chen and colleagues hypothesized that these cells might become over-activated and exhausted in the presence of certain bacteria, which leaves them less capable of killing tumor cells. “We’re trying to figure out what it is that these bacteria are doing to promote inflammation exhaustion,” she said, and one possibility is that they might be producing metabolites.

This type of research takes a lot of work and a long time to arrive at any specific conclusions, she pointed out, and it takes additional effort to apply it to humans because the microbiome is so complex that large numbers of patients are needed to find any specific patterns. There’s still plenty of work to be done, she said, but “step by step, we’ll eventually get there.”

Paul Nicolaus is a freelance writer specializing in science, nature, and health. Learn more at

Load more comments